Research map / Vision-language and multimodal / CLIP and zero-shot
Distribution-Aware Prompt Tuning for Vision-Language Models
Eulrang Cho, Jooyeon Kim, Hyunwoo J Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add inter-dispersion loss on text prompts to CoOp+VPT (ablation) | CoOp+VPT · 11 datasets, 16-shot average | accuracy (%): 81.25 -> 81.33 | Improved |
| add intra-dispersion loss on visual prompts to CoOp+VPT (ablation) | CoOp+VPT · 11 datasets, 16-shot average | accuracy (%): 81.25 -> 81.41 | Improved |
| combine both dispersion losses (full DAPT) (ablation) | CoOp+VPT · 11 datasets, 16-shot average | accuracy (%): 81.25 -> 81.62 | Improved |
| add inter-dispersion loss alone (ablation) | CoOp+VPT · Food101 16-shot | accuracy (%): 86.52 -> 86.42 | Hurt |
| add DAPT dispersion losses with 1 shot per class (ablation) | CoOp+VPT · 11 datasets, 1-shot average | accuracy (%): 61.05 -> 61.42 | Improved |
| use random-sample prototype instead of mean prototype (ablation) | DAPT intra-dispersion loss · 11 datasets, 16-shot average | accuracy (%): 81.62 -> 81.29 | Hurt |
| use random-sample prototype instead of mean prototype (ablation) | DAPT intra-dispersion loss · 11 datasets, 1-shot average | accuracy (%): 61.42 -> 65.08 | Improved |
| apply DAPT instead of CoOp for domain generalization | CLIP ViT-B/16 on ImageNet variants · ImageNet-A | accuracy (%): 48.97 -> 48.74 | Hurt |
| apply DAPT instead of CoOp for domain generalization | CLIP ViT-B/16 on ImageNet variants · ImageNetV2 | accuracy (%): 64.22 -> 64.93 | Improved |
| apply DAPT on base-to-new generalization versus CoOp | CLIP ViT-B/16 · OxfordPets new classes | accuracy (%): 95.43 -> 72.43 | Mixed |
| set inter-dispersion weight beta_t to 100 (ablation) | DAPT · hyperparameter analysis | accuracy | Hurt |
| increase VPT learning rate from 0.002 to 0.2 (ablation) | VPT · 11 datasets, 16-shot average | accuracy (%): 68.32 -> 76.56 | Improved |
Closest papers by meaning
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · similarity 0.93)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.93)
- Make Prompts Adaptable: Bayesian Modeling for Vision-Language Prompt Learning with Data-Dependent Prior (AAAI 2024 · similarity 0.91)
- ZIP: An Efficient Zeroth-order Prompt Tuning for Black-box Vision-Language Models (ICLR 2025 · similarity 0.91)
- Skip Tuning: Pre-trained Vision-Language Models are Effective and Efficient Adapters Themselves (CVPR 2025 · similarity 0.91)
- DPC: Dual-Prompt Collaboration for Tuning Vision-Language Models (CVPR 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
